2013-05-17 08:44:18

by Li Zhong

Subject: [RFC PATCH nohz] return NOTIFY_BAD in cpu down call back to stop offlining the cpu

In tick_nohz_cpu_down_callback() if the cpu is the one handling
timekeeping , it seems that we should return something that could stop
notify CPU_DOWN_PREPARE, and then start notify CPU_DOWN_FAILED on the
already called notifier call backs.

-EINVAL will be converted to 0 by notifier_to_errno(), then the cpu
would be taken down with part of the DOWN_PREPARE notifier callbacks
called, and something bad could happen after that.

Signed-off-by: Li Zhong <[email protected]>
---
kernel/time/tick-sched.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index bc67d42..17b8155 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -306,7 +306,7 @@ static int __cpuinit tick_nohz_cpu_down_callback(struct notifier_block *nfb,
* we can't safely shutdown that CPU.
*/
if (have_nohz_full_mask && tick_do_timer_cpu == cpu)
- return -EINVAL;
+ return NOTIFY_BAD;
break;
}
return NOTIFY_OK;
--
1.7.1

2013-05-20 09:53:44

by Srivatsa S. Bhat

Subject: Re: [RFC PATCH nohz] return NOTIFY_BAD in cpu down call back to stop offlining the cpu

On 05/17/2013 02:14 PM, Li Zhong wrote:
> In tick_nohz_cpu_down_callback() if the cpu is the one handling
> timekeeping , it seems that we should return something that could stop
> notify CPU_DOWN_PREPARE, and then start notify CPU_DOWN_FAILED on the
> already called notifier call backs.
>
> -EINVAL will be converted to 0 by notifier_to_errno(),

This above line is not relevant here, because notifier_call_chain()
doesn't use notifier_to/from_errno(). It simply uses a straight-forward
check like this:

if ((ret & NOTIFY_STOP_MASK) == NOTIFY_STOP_MASK)
break;

> then the cpu
> would be taken down with part of the DOWN_PREPARE notifier callbacks
> called, and something bad could happen after that.
>
> Signed-off-by: Li Zhong <[email protected]>
> ---

Reviewed-by: Srivatsa S. Bhat <[email protected]>

> kernel/time/tick-sched.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
> index bc67d42..17b8155 100644
> --- a/kernel/time/tick-sched.c
> +++ b/kernel/time/tick-sched.c
> @@ -306,7 +306,7 @@ static int __cpuinit tick_nohz_cpu_down_callback(struct notifier_block *nfb,
> * we can't safely shutdown that CPU.
> */
> if (have_nohz_full_mask && tick_do_timer_cpu == cpu)
> - return -EINVAL;
> + return NOTIFY_BAD;
> break;
> }
> return NOTIFY_OK;
>

Regards,
Srivatsa S. Bhat

2013-05-20 10:12:43

by Frederic Weisbecker

Subject: Re: [RFC PATCH nohz] return NOTIFY_BAD in cpu down call back to stop offlining the cpu

2013/5/20 Srivatsa S. Bhat <[email protected]>:
> On 05/17/2013 02:14 PM, Li Zhong wrote:
>> In tick_nohz_cpu_down_callback() if the cpu is the one handling
>> timekeeping , it seems that we should return something that could stop
>> notify CPU_DOWN_PREPARE, and then start notify CPU_DOWN_FAILED on the
>> already called notifier call backs.
>>
>> -EINVAL will be converted to 0 by notifier_to_errno(),
>
> This above line is not relevant here, because notifier_call_chain()
> doesn't use notifier_to/from_errno(). It simply uses a straight-forward
> check like this:
>
> if ((ret & NOTIFY_STOP_MASK) == NOTIFY_STOP_MASK)
> break;
>
>> then the cpu
>> would be taken down with part of the DOWN_PREPARE notifier callbacks
>> called, and something bad could happen after that.
>>
>> Signed-off-by: Li Zhong <[email protected]>
>> ---
>
> Reviewed-by: Srivatsa S. Bhat <[email protected]>

I applied the patch and will send to Ingo, thanks guys!

>
>> kernel/time/tick-sched.c | 2 +-
>> 1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
>> index bc67d42..17b8155 100644
>> --- a/kernel/time/tick-sched.c
>> +++ b/kernel/time/tick-sched.c
>> @@ -306,7 +306,7 @@ static int __cpuinit tick_nohz_cpu_down_callback(struct notifier_block *nfb,
>> * we can't safely shutdown that CPU.
>> */
>> if (have_nohz_full_mask && tick_do_timer_cpu == cpu)
>> - return -EINVAL;
>> + return NOTIFY_BAD;
>> break;
>> }
>> return NOTIFY_OK;
>>
>
> Regards,
> Srivatsa S. Bhat
>